[Digital Today reporter Hyunwoo Chu] The U.S. Department of Defense has signed contracts with Nvidia, Microsoft, Amazon Web Services and Reflection AI to deploy those companies’ AI technologies and models on classified networks. On the 1st (local time), tech outlet TechCrunch reported the move is meant to ensure the department’s use of AI is lawful.
The agreements cover SpaceX, OpenAI, Google, Nvidia, Reflection AI, Microsoft and AWS. Impact Levels 6 and 7 correspond to classified and top-secret operations, so these models will run in environments that contain sensitive intelligence and operational data.
The Pentagon selected multiple U.S. suppliers to avoid single-vendor dependence. It intends to keep both closed models and open-source options available to broaden its choices.
Roles are divided by company. Nvidia will provide models from its NeMoTron open-source family, and Reflection AI will supply additional open-weight systems. Google will adapt its Gemini family for lawful government use. SpaceX is expected to provide infrastructure linked to xAI’s Grok model, while Microsoft and AWS will continue to provide cloud and infrastructure support.
Internal use of AI is already growing. The Defense Department’s generative AI platform, GenAI.mil, exceeded 1.3 million users within five months of launch and logged tens of millions of prompts.
Anthropic was excluded from the list. In February, Defense Secretary Pete Hegseth labeled Anthropic a supply-chain risk, saying the company had not removed limits related to autonomous lethal weapons and large-scale domestic surveillance. A federal judge later blocked enforcement of that ban, and legal disputes remain ongoing.
OpenAI said it will retain its three safety principles under the contract. Other companies accepted the phrase “for all lawful purposes” without attaching similar public conditions.
The move aligns with the Defense Department’s AI acceleration strategy unveiled earlier in 2026. The department plans to scale a modular, open-source architecture across combat, intelligence and administrative functions, prioritize transparent open-weight options from U.S. suppliers, and accelerate prototype development.
Meanwhile, the Defense Department said more than 1.3 million people have used the secure generative AI platform GenAI.mil. The platform runs large language models and other AI tools in a government-approved cloud environment and is primarily used for nonclassified tasks such as research, drafting documents and data analysis.
